Automatic Discoveries of Physical and Semantic Concepts via Association Priors of Neuron Groups

TL;DR

This paper proposes a prior-based method that enables neural networks to automatically associate neurons with physical and semantic concepts across layers, enhancing interpretability and aligning learned representations with real-world concepts.

Contribution

It introduces a novel prior that guides neural networks to naturally learn and associate concepts at different layers without explicit supervision beyond labels.

Findings

Neural networks can learn physical concepts like filter rotation.

Semantic concepts such as fine-grained categories are captured.

The proposed prior improves interpretability of learned representations.

Abstract

The recent successful deep neural networks are largely trained in a supervised manner. It {\it associates} complex patterns of input samples with neurons in the last layer, which form representations of {\it concepts}. In spite of their successes, the properties of complex patterns associated a learned concept remain elusive. In this work, by analyzing how neurons are associated with concepts in supervised networks, we hypothesize that with proper priors to regulate learning, neural networks can automatically associate neurons in the intermediate layers with concepts that are aligned with real world concepts, when trained only with labels that associate concepts with top level neurons, which is a plausible way for unsupervised learning. We develop a prior to verify the hypothesis and experimentally find the proposed prior help neural networks automatically learn both basic physical concepts at the lower layers, e.g., rotation of filters, and highly semantic concepts at the higher layers, e.g., fine-grained categories of an entry-level category.